12 Jan
2024
12 Jan
'24
1:54 p.m.
On Fri Jan 12 13:50:07 2024 +0000, Conor McCarthy wrote:
I doubt we'll ever see a case of a `phi` having only one incoming after deduplication, so an error could be emitted for it, but it could also be handled after structurisation.
I don't think there is any need to emit an error for a PHI with a single incoming block. It's a bit silly, but it should fit any processing algorithm I have in mind without special handling.
OTOH, do you think there is any condition under which we could have a PHI node in which registers have a modifier? Especially for the destination register, it would be a bit annoying for my algorithm which removes PHI nodes.